SAN FRANCISCO — One recent afternoon, the city’s newest grocery market was trying to figure out whether I would buy, steal or leave behind a bag of white cheddar popcorn — and so was I.
On its side: 27 cameras along the ceiling and a wealth of behavioral data.
On my side: crippling indecision.
Last week, San Francisco got its first completely automated cashierless store, Standard Market. Shoppers who have downloaded the store’s app can go into the 1,900-square-foot space, grab items and simply leave. There is no check-in gate, and there is no checkout swipe. Ceiling cameras identify the shopper and the items, and determine when said items leave with said shopper. Or, at least, that’s the idea.
The startup behind this operation is Standard Cognition, which has raised $11.2 million in venture capital and formed partnerships with four retails chains around the world. This first market is a prototype to showcase the technology and work on the bugs. The ambitious goal is to add the tech in 100 stores a day (each day!) by 2020.
Five of the seven founders came from the Securities and Exchange Commission, where they built artificial intelligence software to detect fraud and trade violations, before starting Standard Cognition in 2017. Now these fraud experts are working to discern something equally complicated: whether I am stealing a snack.
Standard Market is the latest entry in the emerging fray of retail automation, where companies are throwing cameras, sensors and machine learning into grocery stores to replace the checkout line. In January, Amazon opened its first cashierless Go market in Seattle to the public; it has since opened more of the stores. In China, experiments in cashierless stores abound, using radio frequency identification tags and a self-checkout process that involves scanning a Quick Response code or your face.
Standard Cognition’s approach is different. It relies exclusively on the ceiling cameras and artificial intelligence software to figure out what you are buying. The cameras document shoppers’ movements, speed, stride length and gaze. The store knows when I glance at a poster and for how long. It knows if I slowed down, grabbed a chocolate bar and put it back. It knows if my body is facing the dried mangoes but my face is set on the popcorn.
And it knows (or is trying to know) when I am planning to steal.
The goal is to predict, and prevent, shoplifting, because unlike Amazon’s Go stores, which have a subway turnstile-like gate for entry and exit, Standard Market has an open door, and the path is clear.
“We learn behaviors of what it looks like to leave,” said Michael Suswal, Standard Cognition’s co-founder and chief operating officer. Trajectory, gaze and speed are especially useful for detecting theft, he said, adding, “If they’re going to steal, their gait is larger, and they’re looking at the door.”
Once the system decides it has detected potential theft behavior, a store attendant will get a text and walk over for “a polite conversation,” Suswal said.
Predicting theft requires a lot of data about shoppers, much of which does not exist yet — “or at least no one is willing to give it to us,” he said.
So a few days before Standard Market opened, Standard Cognition hired 100 actors to shop there for four hours. In Japan, the team has worked with a convenience store chain, whose name it has not disclosed, in a very useful data collection effort.